Autoencoder
In plain terms
A network that compresses data then reconstructs it.
Going deeper
An autoencoder learns to encode input into a compact representation and decode it back, useful for compression, denoising, and feature learning. It learns efficient codes.
Under the hood
The bottleneck forces useful representations; variational autoencoders enable generation.
Examples
Anomaly detection, denoising, dimensionality reduction, VAEs, embeddings.
Think of it like...
Like summarizing a book to its essence, then rewriting it from the summary.